What the work involves

You receive model output that looks like accounting work: a proposed journal entry with an explanation, a bank or intercompany reconciliation, a revenue recognition memo, a variance walk, a close workpaper. Your job is to decide whether it is technically correct under the relevant standard, then say precisely where it goes wrong and what the correct treatment is. Much of the day is annotation and rubric-based scoring — rating or ranking two or more responses, tagging the specific error (wrong sign, wrong period, wrong account, plausible-sounding but fabricated standard citation), and writing rationale a model team can act on. You will also be asked to author original scenarios and reference answers drawn from how close and reporting actually run in NetSuite, SAP, Oracle, QuickBooks, Xero, or Dynamics 365 — not textbook exercises.

What the platform screens for

micro1's screen is AI-led and leans on follow-up questions. Two gates dominate: substantive hands-on accounting (three years or more in close, corporate accounting, audit, bookkeeping, reporting, cost accounting, or FP&A with real accounting content), and prior paid human-data experience for AI — labeling, annotation, RLHF, response evaluation, model evaluation, or rubric grading. The listing is explicit that the second is a firm requirement and that AI features inside accounting software do not satisfy it, so expect to name the program, the task type, and the pay arrangement. Beyond credentials, the screen probes whether you can hold a rubric steady: whether you distinguish a wrong answer from a poorly explained right answer, whether you escalate ambiguity rather than inventing a house rule, and whether you can write feedback that is specific enough to be actionable.

Logistics

  • Fully remote contractor engagement, asynchronous batch work with rubric documents and calibration rounds.
  • Observed pay band $90–170/hr, varying with project track and depth of systems experience; rates are as observed on the platform and not guaranteed.
  • Volume is project-driven and can be uneven; the listing states immediate availability and a reliable connection are expected.
  • Written English at B2 or above, since most of the deliverable is written rationale.

Who tends to do well

Controllers, senior accountants, audit seniors and managers, and FP&A analysts who have already done a stint of paid annotation or evaluation work. Depth in one system plus fluency in Excel beats a shallow list of eight ERPs, and being able to say what actually breaks at close in your system carries more weight than certification names.